Information generation method and device, terminal, cloud server and medium

Abstract

The invention relates to an information generation method and device, a terminal, a cloud server and a medium. The method comprises the steps of generating first information by adopting a first language model based on first input information, and obtaining an evaluation parameter corresponding to the first information; the evaluation parameter is determined by the cloud server, the evaluation parameter is used for representing the association degree of the first information and the first input information, and when the evaluation parameter is greater than a preset threshold value, the first information is determined as first output information corresponding to the first input information. According to the information processing method and device, the first information can be evaluated through the cloud server, the accuracy of the first information can be determined, then when the evaluation parameter corresponding to the first information is larger than the preset threshold value, the first information is determined as the output information, and therefore the accuracy of the output information is guaranteed. Thus, the cloud server can cooperate with the terminal to complete the information generation process, the economic cost can be saved, the quality and accuracy of the output information can be improved, and then the user experience is improved.

Description

Technical Field

[0001] This disclosure relates to the field of terminal technology, and in particular to an information generation method, apparatus, terminal, cloud server and medium. Background Technology

[0002] A large language model is a powerful language model that can construct coherent information through continuously generated semantic units during the reasoning stage.

[0003] Currently, large language models can be configured either on the terminal or on a cloud server. However, neither configuration on the terminal nor on a cloud server provides a good user experience. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this disclosure provides an information generation method, apparatus, terminal, cloud server, and medium.

[0005] According to a first aspect of the present disclosure, an information generation method is provided, comprising:

[0006] Based on the first input information, the first information is generated using a first language model; the first language model is a large language model configured in the terminal.

[0007] Obtain the evaluation parameters corresponding to the first information; the evaluation parameters are determined by the cloud server and are used to characterize the degree of correlation between the first information and the first input information.

[0008] When the evaluation parameter is greater than a preset threshold, the first information is determined as the first output information corresponding to the first input information.

[0009] In some embodiments, the method further includes:

[0010] When the evaluation parameter is less than or equal to the preset threshold, second information is obtained; the second information is information generated by the second language model based on the first input information, and the second language model is the large language model configured in the cloud server.

[0011] The second information is determined as the first output information corresponding to the first input information.

[0012] In some embodiments, the first input information includes initial input information and historical output information, the historical output information including output information generated based on the initial input information prior to the first information, and the method further includes:

[0013] If no target identifier is detected from the first output information, the initial input information, the historical output information, and the first output information are determined as the second input information; the target identifier is used to indicate the end of the information generation process.

[0014] Based on the second input information, determine the second output information corresponding to the second input information until the target identifier is detected from the output information.

[0015] In some embodiments, the method further includes:

[0016] If the target identifier is detected from the first output information, the historical output information and the first output information are determined as the target output information corresponding to the initial input information.

[0017] In some embodiments, the method further includes:

[0018] Obtain updated initial input information; the updated initial input information includes the initial input information and the prompt information corresponding to the initial input information, the prompt information being used to characterize the angle of answering the question described by the initial input information;

[0019] Based on the updated initial input information, the first input information is determined.

[0020] According to a second aspect of the present disclosure, an information generation method is provided, comprising:

[0021] Obtain first information, which is information generated by a first language model based on first input information, and the first language model is a large language model configured in the terminal;

[0022] The first information and the first input information are processed to obtain the evaluation parameters corresponding to the first information; the evaluation parameters are used to characterize the degree of correlation between the first information and the first input information.

[0023] In some embodiments, the first information includes a plurality of semantic units, and the processing of the first information and the first input information to obtain the evaluation parameters corresponding to the first information includes:

[0024] The first input information and each semantic unit are processed to obtain the evaluation parameters of each semantic unit;

[0025] The evaluation parameters corresponding to the first information are determined based on the evaluation parameters of each semantic unit.

[0026] In some embodiments, processing the first information and the first input information to obtain the evaluation parameters corresponding to the first information includes:

[0027] Based on the evaluation model, the first information and the first input information are processed to obtain the evaluation parameters corresponding to the first information; the evaluation model is used to evaluate the degree of correlation between the two types of input information.

[0028] In some embodiments, the method further includes:

[0029] When the evaluation parameter is less than or equal to a preset threshold, second information is generated using the second language model based on the first input information.

[0030] In some embodiments, the first input information includes initial input information, and the method further includes:

[0031] Determine prompt information corresponding to the initial input information, wherein the prompt information is used to characterize the angle of answering the question described by the initial input information;

[0032] The prompt information is concatenated with the initial input information to obtain the updated initial input information.

[0033] According to a third aspect of the present disclosure, an information generation apparatus is provided, comprising:

[0034] The generation module is configured to generate first information based on the first input information and using a first language model; the first language model is a large language model configured in the terminal.

[0035] The first acquisition module is configured to acquire the evaluation parameters corresponding to the first information; the evaluation parameters are determined by the cloud server and are used to characterize the degree of correlation between the first information and the first input information.

[0036] The determination module is configured to determine the first information as the first output information corresponding to the first input information when the evaluation parameter is greater than a preset threshold.

[0037] According to a fourth aspect of the present disclosure, an information generation apparatus is provided, comprising:

[0038] The second acquisition module is configured to acquire first information, which is information generated by a first language model based on the first input information, and the first language model is a large language model configured in the terminal.

[0039] The processing module is configured to process the first information and the first input information to obtain the evaluation parameters corresponding to the first information; the evaluation parameters are used to characterize the degree of correlation between the first information and the first input information.

[0040] According to a fifth aspect of the present disclosure, a terminal is provided, comprising:

[0041] processor;

[0042] Memory used to store processor-executable instructions;

[0043] The processor is configured to perform the information generation method as described in the first aspect of this disclosure.

[0044] According to a sixth aspect of the present disclosure, a cloud server is provided, comprising:

[0045] processor;

[0046] Memory used to store processor-executable instructions;

[0047] The processor is configured to perform the information generation method as described in the second aspect of this disclosure.

[0048] According to a seventh aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, which, when instructions in the storage medium are executed by a processor of a terminal, enables the terminal to perform the information generation method as described in the first aspect of the present disclosure.

[0049] According to an eighth aspect of the present disclosure, a non-transitory computer-readable storage medium is provided, which, when instructions in the storage medium are executed by a processor of a cloud server, enables the cloud server to perform the information generation method as described in the second aspect of the present disclosure.

[0050] The method described in this disclosure has the following beneficial effects: This disclosure allows the first information to be evaluated by a cloud server, determining its accuracy. Then, when the evaluation parameter corresponding to the first information exceeds a preset threshold, the first information is determined as the output information, thereby ensuring the accuracy of the output information. In this way, the cloud server can collaborate with the terminal to complete the information generation process, which not only saves economic costs but also improves the quality and accuracy of the output information, thereby enhancing the user experience.

[0051] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0052] The accompanying drawings, which are incorporated in and form a part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure.

[0053] Figure 1 This is a flowchart illustrating an information generation method according to an exemplary embodiment.

[0054] Figure 2 This is a flowchart illustrating an information generation method according to an exemplary embodiment.

[0055] Figure 3 This is a flowchart illustrating an information generation method according to an exemplary embodiment.

[0056] Figure 4 This is a flowchart illustrating an information generation method according to an exemplary embodiment.

[0057] Figure 5 This is a flowchart illustrating an information generation method according to an exemplary embodiment.

[0058] Figure 6 This is a block diagram illustrating an information generation apparatus according to an exemplary embodiment.

[0059] Figure 7 This is a block diagram illustrating an information generation apparatus according to an exemplary embodiment.

[0060] Figure 8 This is a block diagram of a terminal according to an exemplary embodiment. Detailed Implementation

[0061] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this disclosure. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this disclosure as detailed in the appended claims.

[0062] Large language models (MLMs) are powerful language models that construct coherent information through iteratively generated semantic units during the inference phase. In the inference phase, the user provides initial input information, which can be a sentence, a question, or any other form of input, to initiate the MLM's generation process. Upon receiving the initial input, the MLM continuously generates semantic units based on language patterns learned during pre-training, until the number of generated semantic units reaches a limit or a target identifier is generated. The target identifier represents the end of the information generation process. Finally, the MLM forms the target output information based on all the semantic units generated from the initial input information; this target output information is the response corresponding to the initial input information.

[0063] Currently, large language models can be configured either on a terminal or on a cloud server. When configured on a terminal, the inference phase can be performed on the device itself, eliminating the need to transmit input information to the cloud server. This reduces data transmission time and enables rapid response to initial input. Furthermore, since the inference phase is completed on the terminal, there are no cloud server usage and maintenance costs, further reducing the economic cost of using the large language model. However, a large language model configured on a terminal may be limited by the device's performance, potentially resulting in lower quality or accuracy of output information generated from the input, leading to a poor user experience. Conversely, when configured on a cloud server, the vast computing resources of the cloud provider allow for the generation of higher quality and more accurate output information from the initial input. However, network data transmission between the cloud server and the terminal requires a longer response time. Additionally, the use and maintenance of cloud servers incur higher economic costs and may also result in a poor user experience.

[0064] To address the aforementioned problems, this disclosure provides an information generation method. This method includes generating first information based on first input information using a first language model, and obtaining evaluation parameters corresponding to the first information. The evaluation parameters are determined by a cloud server. When the evaluation parameters exceed a preset threshold, the first information is identified as the first output information corresponding to the first input information. This disclosure allows the cloud server to evaluate the first information, determining its accuracy. Then, when the evaluation parameters corresponding to the first information exceed the preset threshold, the first information is identified as the output information, thereby ensuring the accuracy of the output information. In this way, the cloud server can collaborate with the terminal to complete the information generation process, saving economic costs and improving the quality and accuracy of the output information, thus enhancing the user experience.

[0065] The information generation method provided in this embodiment is executed by a terminal, which can be a smart device such as a mobile phone, tablet computer, laptop, smart robot, or smart wearable device. In addition, the terminal is equipped with various hardware resources and an energy storage device that provides power for the operation of these hardware resources.

[0066] Figure 1 This is a flowchart illustrating an information generation method according to an exemplary embodiment, executed by a terminal. See also... Figure 1 The method includes the following steps:

[0067] Step S101: Based on the first input information, generate first information using a first language model; the first language model is the large language model configured in the terminal.

[0068] The first input information includes initial input information and historical output information. The initial input information can be text information comprising multiple bytes, or it can be voice information input by the user. The nature of the initial input information can be various forms of input, such as questions, descriptions, or topics. For example, the initial input information could be "Please briefly describe virtual functions in C++". Historical output information includes output information generated based on the initial input information prior to the first information. For example, historical output information could be "In C++, a virtual function is a special type of member function".

[0069] The first information comprises multiple semantic units, which are the basic data units. Semantic units can be words, letters, numbers, punctuation marks, characters, etc. The first language model is a pre-trained large language model configured in the terminal. The first language model can generate the first information corresponding to the first input information based on it.

[0070] In some embodiments, to obtain the target output information, multiple information generation processes need to be completed based on the input information, and finally, the output information from each information generation process is merged into the target output information. For example, obtaining the target output information based on the initial input information requires three information generation processes: the first information generation process generates output information A, the second information generation process generates output information B, and the third information generation process generates output information D. Therefore, the target output information includes output information A, output information B, and output information D. Historical output information refers to the output information from all information generation processes based on the initial input information that precede the current information generation process. The current information generation process refers to one of the multiple information generation processes.

[0071] To facilitate understanding, the following example uses three information generation processes. In the first information generation process, the initial input information can be directly used as input to generate output information 1. In the second information generation process, the initial input information and output information 1 can be used as input to generate output information 2. At this time, output information 1 is the historical output information. In the third information generation process, the initial input information, output information 1, and output information 2 are used as output to generate output information 3. At this time, output information 1 and output information 2 are the historical output information.

[0072] It should be noted that when the information generation process is the first information generation process, the first input information only includes the initial input information.

[0073] In some embodiments, the number of semantic units in the output information during each information generation process can be preset, for example, setting the number of semantic units to 20, 30, etc. It should be noted that the number of semantic units in the output information during each information generation process can be the same or different. For example, the number of semantic units in the output information corresponding to the first and second information generation processes can both be 20; or, for example, the number of semantic units in the output information corresponding to the first information generation process can be 20, and the number of semantic units in the output information corresponding to the second information generation process can be 10. In some embodiments, the number of semantic units in the output information is related to the quality of the output information. Specifically, the fewer semantic units included in the output information, the higher the quality of the output information; conversely, the more semantic units included in the output information, the lower the quality of the output information. For ease of understanding, taking the output information as the first information as an example, the relationship between the number of semantic units in the output information and the quality of the output information is explained as follows:

[0074] When the initial information contains a small number of semantic units, more information generation processes are required to obtain the final output information. Therefore, the quality of the output information obtained through more information generation processes will be higher. However, when the initial information contains a small number of semantic units, more information generation processes also mean that the total information generation process takes more time, and consequently, the efficiency of the information generation process may be lower. Conversely, when the initial information contains a large number of semantic units, fewer information generation processes are required than when the initial information contains a small number of semantic units. Therefore, the quality of the output information may be lower, but the efficiency of the information generation process is higher.

[0075] In some embodiments, based on the example shown in the first information above, the number of semantic units included in the output information can be limited according to actual needs. If a faster information generation process is required, the output information can be set to include a larger number of semantic units; conversely, if higher quality output information is required, the output information can be set to include a smaller number of semantic units. During application, a suitable number can be determined to simultaneously meet the requirements of high quality and high efficiency, thereby improving the user experience in both aspects.

[0076] In some embodiments, the user needs represented by the first input information may not be explicit. Therefore, in order to further clarify the user needs represented by the first input information and obtain more accurate output information, the initial input information can be updated, and the first input information can be determined based on the updated initial input information. Then, based on the first input information, the first information can be generated using a first language model.

[0077] In some embodiments, the updated initial input information may include the initial input information and corresponding prompts. The prompts characterize the perspective from which the initial input information addresses the question, where the perspective can be represented as a prefix or suffix of the initial input information (e.g., in the field of E, as an expert in the field of E). For example, if the initial input information is "virtual functions in C++", the updated initial input information might be "As a computer expert, please briefly describe virtual functions in C++". Additionally, the prompts can also be used to characterize the information related to the planned initial input information. Specifically, the prompts can be represented as a planned template, such as: "I want to know some information / explanation about (initial input information)," "I need some examples / cases about (initial input information)," "Please help me understand the principles / mechanisms behind (initial input information)," or "Please help me understand the main characteristics of (initial input information)." Furthermore, when the prompts are used to characterize the information related to the planned initial input information, keywords from the initial input information can be added to the template. The updated initial input information has the same meaning as the initial input information. The updated initial input information may be a modified version of the initial input information. For example, the updated initial input information may be a normalized version of the initial input information. For example, if the initial input information is "virtual functions in C++", the updated initial input information may be "please explain the meaning and function of virtual functions in C++".

[0078] The updated initial input information can more accurately represent user needs, enabling the first language model to generate more accurate initial information and thus improving the accuracy of the output information. Updating the initial input information can be performed by a cloud server, which pre-stores multiple prompts. The cloud server can determine the appropriate prompt based on the initial input information, and then determine the updated initial input information based on both the prompts and the original initial input information.

[0079] Step S102: Obtain the evaluation parameters corresponding to the first information; the evaluation parameters are determined by the cloud server.

[0080] The evaluation parameter characterizes the degree of correlation between the first information and the first input information. Specifically, the degree of correlation between the first information and the first input information can be considered as the correlation between them. The greater the correlation between the first information and the first input information, the larger the value of the evaluation parameter, and the more accurate the first information. For example, when the first input information is "you," if the first information is "good," the correlation between the first information and the first input information is high, and the value of the evaluation parameter is large; conversely, if the first information is "sun," the correlation between the first information and the first input information is low, and the value of the evaluation parameter is small.

[0081] In some embodiments, the evaluation parameters can be represented numerically, thereby allowing the accuracy of the first information to be determined intuitively. The evaluation parameters can take values ​​in the range of [0, 1], for example, the evaluation parameters can be 0.6, 0.8, etc., or [0, 100].

[0082] In some embodiments, before generating the first information using the first language model, it is necessary to determine the current network status of the terminal. If there are problems with the terminal's current network status, such as no network connection, network lag, or network congestion, the large language model configured in the terminal, i.e., the first language model, is directly used to complete all information generation processes. If there are no problems with the terminal's current network status, such as a smooth network, the cloud server can participate in the information generation process. The cloud server and the terminal collaborate to complete the information generation process, thereby achieving end-to-cloud collaboration, saving economic costs and improving the accuracy of the output information.

[0083] In some embodiments, after determining that there are no problems with the current network status and after generating the first information, the terminal uploads the first information to the cloud server via data transmission. Specifically, uploading the first information to the cloud server can be achieved through various types of data transmission protocols, such as FTP (File Transfer Protocol), where FTP client software (e.g., FileZilla) can be used to connect to the cloud server and upload information; SCP (Secure Copy Protocol), where the SCP command can be used to copy information from the terminal to the cloud server; SFTP (Secure File Transfer Protocol), where SFTP client software (e.g., FileZilla) can be used to connect to the cloud server and upload information; additionally, command-line tools can also be used to upload information to the cloud server.

[0084] Subsequently, the evaluation parameters can be determined by the cloud server. Specific evaluation parameters can be determined by the evaluation model configured on the cloud server, which is used to evaluate the degree of correlation between the first input information and the first information. It should be noted that the large language model can be applied to text classification scenarios; therefore, the evaluation model can be a second language model, which is the large language model configured on the cloud server. Alternatively, the evaluation model can also be a model distinct from the second language model. When the evaluation model is distinct from the second language model, the training data for the evaluation model includes multiple sets of training samples and the evaluation parameters corresponding to each set of training samples. Each set of training samples includes the input information and corresponding output information of the second language model.

[0085] By evaluating the first information through a cloud server, the first information can be made close to the output information of the cloud server based on the first input information, thereby ensuring the accuracy and quality of the first information.

[0086] Step S103: When the evaluation parameter is greater than the preset threshold, the first information is determined as the first output information corresponding to the first input information.

[0087] The preset threshold can be set according to actual needs. If you want to get higher quality and more accurate output information, you can set the value of the preset threshold to be larger. For example, when the evaluation parameter range is [0, 1], the value of the preset threshold can be set to 0.8.

[0088] In some embodiments, when the evaluation parameter is greater than a preset threshold, it indicates that the first information is close to the output information of the cloud server based on the first input information, that is, the accuracy and quality of the first information are high. Therefore, the first information can be determined as the first output information corresponding to the first input information.

[0089] Step S104: When the evaluation parameter is less than or equal to a preset threshold, obtain the second information and determine the second information as the first output information corresponding to the first input information; the second information is the information generated by the second language model based on the first input information.

[0090] If the evaluation parameter is less than or equal to a preset threshold, it indicates that the first information is not close to the output information of the cloud server based on the first input information, and the accuracy and quality of the first information are low. Therefore, the first information cannot be used as the first output information corresponding to the first input information. In some embodiments, when the quality or accuracy of the first information output by the first language model configured in the terminal does not meet the requirements, the second information corresponding to the first input information can be determined from the second language model configured in the cloud server, and the second information can be determined as the first output information.

[0091] In this embodiment, the accuracy of the first information is determined by evaluating it through a cloud server. Then, when the evaluation parameter corresponding to the first information exceeds a preset threshold, the first information is designated as the output information, thus ensuring the accuracy of the output information. In this way, the cloud server can collaborate with the terminal to complete the information generation process, which not only saves economic costs but also improves the quality and accuracy of the output information, thereby enhancing the user experience.

[0092] In some embodiments, steps S101-S104 above describe the information generation process of one output information. In applications, the language model cannot generate all the output information in one information generation process. Therefore, in practice, it is necessary to continuously cycle the information generation process to obtain the final output information, i.e., the target output information. The following will illustrate... Figure 2 The illustrated embodiment explains the cyclical process of information generation.

[0093] Figure 2 This is a flowchart illustrating an information generation method according to an exemplary embodiment, executed by a terminal. See also... Figure 2 The method includes the following steps:

[0094] Step S201: If no target identifier is detected from the first output information, the initial input information, historical output information and the first output information are determined as the second input information.

[0095] The target identifier is used to characterize the end of the information generation process. In some embodiments, the target identifier can be a symbol indicating the end of the information generation process, such as a period. Alternatively, the target identifier can also be a limit on the number of output semantic units; for example, when the number of output semantic units is x, it is determined that the target identifier has been detected.

[0096] If the target identifier is not detected in the first output information, it indicates that the information generation process is not yet complete and the information generation loop needs to continue. It should be noted that during the loop, the input information needs to be continuously updated to improve the correlation between the input and output information. Therefore, after outputting the first output information, the initial input information, historical output information, and the first output information can be merged, and the merged information can be used as the second input information.

[0097] Step S202: Based on the second input information, determine the second output information corresponding to the second input information until the target identifier is detected from the output information.

[0098] The process of determining the second output information corresponding to the second input information based on the second input information includes:

[0099] Based on the second input information, the first language model is used to generate third information, and the evaluation parameters corresponding to the third information are obtained. When the evaluation parameters are greater than a preset threshold, the third information is determined as the second output information; otherwise, the information generated by the second language model based on the second input information is determined, and this information is determined as the second output information. The process of determining the second output information is the same as steps S101-S104.

[0100] After determining the second output information, it can be determined whether a target identifier is detected in the second output information. If yes, step S203 is executed; if no, the second output information is used as historical output information, and steps S201 and S202 are executed continuously to determine the output information until a target identifier is detected.

[0101] Step S203: If a target identifier is detected from the first output information, the historical output information and the first output information are determined as the target output information corresponding to the initial input information.

[0102] If a target identifier is detected in the first output information, it means that the information generation process has ended. Therefore, the historical output information and the current first output information can be identified as the target output information corresponding to the initial input information.

[0103] In some embodiments, initial input information and target output information can be output simultaneously and displayed on the terminal's screen. It should be noted that when displaying initial input information and target output information, it is necessary to distinguish between them, for example, by using an input identifier. <prompt>and output identifier <answer>The initial input information and the target output information are distinguished separately, and can be represented as follows: <prompt>As a computer expert, please briefly describe virtual functions in C++. <answer>In C++, a virtual function is a special type of member function that can be overridden by subclasses and dynamically bound to the correct function implementation at runtime. Virtual functions are defined by adding the keyword "virtual" before the function declaration. When a function in a class is declared as virtual, a pointer to a virtual function table is added to the object's memory layout. This pointer points to a virtual function table that stores the addresses of all virtual functions in the class and its parent classes.

[0104] In this embodiment of the disclosure, by continuously executing the information generation process, the final output information and the target output information can be obtained, thereby determining the response corresponding to the initial input information, thus meeting user needs and improving the user experience.

[0105] The information generation method provided in this disclosure is executed by a cloud server, which is a simple, efficient, secure, reliable, and elastically scalable computing service. Cloud servers have strong computing capabilities and can be applied to scenarios such as deep learning and scientific computing.

[0106] Figure 3 This is a flowchart illustrating an information generation method according to an exemplary embodiment, executed by a cloud server. See also... Figure 3 The method includes the following steps:

[0107] Step S301: Obtain the first information.

[0108] Here, the first information is information generated by the first language model based on the first input information, and the first language model is the large language model configured in the terminal. For a description of the first language model, please refer to the content shown in step S101, which will not be repeated here.

[0109] After the terminal uploads the initial information to the cloud server, the cloud server can store the initial information in the data center operated by the cloud service provider. When the initial information is needed, it can be retrieved directly from the data center.

[0110] Step S302: Process the first information and the first input information to obtain the evaluation parameters corresponding to the first information.

[0111] The evaluation parameters are used to characterize the degree of correlation between the first information and the first input information.

[0112] In some embodiments, the first information includes multiple semantic units. The first input information and each semantic unit can be processed to obtain evaluation parameters for each semantic unit. Then, based on the evaluation parameters of each semantic unit, the evaluation parameters corresponding to the first information are determined. It should be noted that the evaluation parameters of each semantic unit can be determined by the degree of association between the first information and the first input information; that is, the evaluation parameters of the semantic unit are determined based on the degree of association between each semantic unit and the first input information. Furthermore, there are also associations between each semantic unit. Therefore, the evaluation parameters of each semantic unit can also be determined jointly by the degree of association between each semantic unit and the first input information, and the degree of association between each semantic unit and the previous semantic unit. This embodiment does not limit the specific method for determining the evaluation parameters of each semantic unit; the specific method for determining the evaluation parameters of each semantic unit can be set and selected based on actual needs.

[0113] In some embodiments, the average value of the evaluation parameters of all semantic units can be calculated and the average value can be determined as the evaluation parameter corresponding to the first information; or, the evaluation parameters of all semantic units can be weighted to obtain a weighted value and the weighted value can be determined as the evaluation parameter corresponding to the first information.

[0114] In some embodiments, the first information and the first input information can be processed based on an evaluation model. Specifically, the first information can be evaluated based on the evaluation model and the first input information to obtain evaluation parameters for the first information. The evaluation model is used to evaluate the degree of correlation between the first input information and the first information. Specifically, the evaluation model can be a second language model, which is a large language model configured in a cloud server. Alternatively, the evaluation model can be a model different from the second language model. When the evaluation model is different from the second language model, the training data of the evaluation model includes multiple sets of training samples and evaluation parameters corresponding to each set of training samples. Each set of training samples includes the input information of the second language model and the output information corresponding to that input information. Thus, when processing the first information and the first input information, the first information can be evaluated based on the information of the second language model, so that the first information corresponding to the evaluation parameters being greater than a preset threshold is close to the information output by the second language model based on the first input information.

[0115] Step S303: When the evaluation parameter is less than or equal to a preset threshold, second information is generated using a second language model based on the first input information.

[0116] When the evaluation parameter is less than or equal to a preset threshold, it indicates that the quality or accuracy of the first information generated by the first language model based on the first input information does not meet the requirements. Therefore, the first information cannot be used as the output information of this information generation process. Since cloud servers can utilize the large-scale computing resources of cloud service providers, and cloud servers have strong computing power, supporting more complex large language models and high-quality output information, a second language model can be used to generate second information to replace the first information.

[0117] The process of generating second information using the second language model can be referred to in step S101, where the first language model generates first information, and will not be repeated here.

[0118] In this embodiment, the cloud server can process the acquired first information and first input information to obtain evaluation parameters corresponding to the first information. This embodiment can determine the evaluation parameters corresponding to the first information, thereby determining whether the first information can be used as the first output information. If the first information cannot be used as the first output information, a second language model is used to generate the first output information. This ensures the accuracy of the first output information and the final output information, thereby improving the user experience.

[0119] In some embodiments, the initial input information may not accurately represent the user's intent, therefore the initial input information can be updated to obtain more accurate output information. The following describes... Figure 4 The illustrated embodiment explains the process of updating the initial input information.

[0120] Figure 4 This is a flowchart illustrating an information generation method according to an exemplary embodiment, executed by a cloud server. See also... Figure 4 The method includes the following steps:

[0121] Step S401: Determine the prompt information corresponding to the initial input information.

[0122] The prompt information is used to characterize the perspective from which the initial input information describes the question. For details regarding the prompt information, please refer to step S101, which will not be repeated here.

[0123] In some embodiments, the cloud server pre-stores multiple prompt messages. The cloud server can obtain the initial input information, determine the appropriate prompt message based on the initial input information, and then determine the updated initial input information based on the prompt message and the initial input information.

[0124] In some embodiments, the corresponding prompt information can be determined based on keywords in the initial input information. Optionally, there is a correspondence between keywords and prompt information. For example, when the keyword in the initial input information is "virtual functions in C++", the prompt information corresponding to the initial input information can be determined to be "as a computer expert" based on the correspondence.

[0125] In some embodiments, the corresponding prompt information can also be determined based on the semantics of the initial input information. For example, when the semantics of the initial input information is to explain a certain term, the corresponding prompt information could be "Please explain the meaning and function of the initial input information (or the keywords in the initial input information) to me".

[0126] Step S402: Concatenate the prompt information with the initial input information to obtain the updated initial input information.

[0127] When the prompt message is a prefix or suffix of the initial input information, it can be directly merged with the initial input information. For example, if the initial input information is "Please briefly describe virtual functions in C++" and the prompt message is the prefix "As a computer expert," the updated initial input information would be: "As a computer expert, please briefly describe virtual functions in C++." Alternatively, when the prompt message is a structured template, the initial input information or keywords from the initial input information can be directly added to the corresponding position in the structured template. For example, if the initial input information is "Please briefly describe virtual functions in C++" and the prompt message is "Please explain the meaning and function of the initial input information (or keywords from the initial input information)," the updated initial input information would be: "Please explain the meaning and function of virtual functions in C++."

[0128] By updating the initial input information and then determining the first input information based on the updated initial input information, the accuracy of the input information can be improved, making the input information more in line with user needs, thereby obtaining more accurate output information and improving the user experience.

[0129] The following is combined with, for example Figure 5 The flowchart of the information generation method is shown, and the interaction process between the cloud server and the terminal is discussed:

[0130] S501, The cloud server determines the prompt information corresponding to the initial input information, and concatenates the prompt information with the initial input information to obtain the updated initial input information.

[0131] S502, the cloud server determines the first input information based on the updated initial input information.

[0132] S503, the cloud server sends the first input information to the terminal.

[0133] S504. The terminal obtains the first input information and generates the first information based on the first input information using the first language model.

[0134] S505. The terminal determines whether the current network status is normal; if yes, proceed to step S506; if no, proceed to step S521.

[0135] S506, The terminal sends the first information to the cloud server.

[0136] S507: The cloud server obtains the first information, processes the first information and the first input information, and obtains the evaluation parameters corresponding to the first information.

[0137] S508, the cloud server sends the evaluation parameters corresponding to the first information to the terminal.

[0138] S509, The terminal obtains the evaluation parameters corresponding to the first information.

[0139] S510. The terminal determines whether the evaluation parameter is greater than the preset threshold; if yes, then proceed to step S511; if no, then proceed to step S512.

[0140] S511, The terminal determines the first information as the first output information corresponding to the first input information.

[0141] S512, the cloud server generates second information based on the first input information using a second language model.

[0142] S513, the cloud server sends the second information to the terminal.

[0143] S514. The terminal obtains the second information and determines the second information as the first output information corresponding to the first input information.

[0144] S515. The terminal determines whether a target identifier has been detected from the first output information; if yes, then proceed to step S516; if no, then proceed to step S517.

[0145] S516. The terminal determines the historical output information and the first output information as the target output information corresponding to the initial input information.

[0146] S517. The terminal determines the initial input information, historical output information and first output information as the second input information.

[0147] S518. The terminal determines the second output information corresponding to the second input information based on the second input information until the target identifier is detected from the output information.

[0148] S519. The terminal determines the historical output information and the output information of the last information generation process as the target output information corresponding to the initial input information.

[0149] S520, the terminal outputs initial input information and target output information.

[0150] S521, The terminal uses a first language model to determine the target output information.

[0151] Figure 6 This is a block diagram illustrating an information generation device configured on a terminal according to an exemplary embodiment. See also... Figure 6 The device includes:

[0152] The generation module 601 is configured to generate first information based on the first input information and using a first language model; the first language model is a large language model configured in the terminal.

[0153] The first acquisition module 602 is configured to acquire the evaluation parameters corresponding to the first information; the evaluation parameters are determined by the cloud server and are used to characterize the degree of correlation between the first information and the first input information.

[0154] The determination module 603 is configured to determine the first information as the first output information corresponding to the first input information when the evaluation parameter is greater than a preset threshold.

[0155] In some embodiments, the determining module 603 is configured to:

[0156] When the evaluation parameter is less than or equal to the preset threshold, the second information is obtained; the second information is the information generated by the second language model based on the first input information, and the second language model is the large language model configured in the cloud server.

[0157] The second information is determined as the first output information corresponding to the first input information.

[0158] In some embodiments, the first input information includes initial input information and historical output information, wherein the historical output information includes output information generated based on the initial input information prior to the first information, and the determining module 603 is configured to:

[0159] If the target identifier is not detected in the first output information, the initial input information, historical output information, and the first output information are determined as the second input information; the target identifier is used to indicate the end of the information generation process.

[0160] Based on the second input information, determine the second output information corresponding to the second input information until the target identifier is detected from the output information.

[0161] In some embodiments, the determining module 603 is configured to:

[0162] If a target identifier is detected from the first output information, the historical output information and the first output information are determined as the target output information corresponding to the initial input information.

[0163] In some embodiments, the determining module 603 is configured to:

[0164] Obtain the updated initial input information; the updated initial input information includes the initial input information and the corresponding prompt information, which is used to characterize the perspective of answering the question described by the initial input information.

[0165] Based on the updated initial input information, determine the first input information.

[0166] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0167] Figure 7 This is a block diagram illustrating an information generation apparatus according to an exemplary embodiment, configured on a cloud server. See also... Figure 7 The device includes:

[0168] The second acquisition module 701 is configured to acquire first information, which is information generated by the first language model based on the first input information, and the first language model is a large language model configured in the terminal.

[0169] The processing module 702 is configured to process the first information and the first input information to obtain the evaluation parameters corresponding to the first information; the evaluation parameters are used to characterize the degree of correlation between the first information and the first input information.

[0170] In some embodiments, the first information includes multiple semantic units, and the processing module 702 is configured to:

[0171] The first input information and each semantic unit are processed to obtain the evaluation parameters for each semantic unit;

[0172] Based on the evaluation parameters of each semantic unit, the evaluation parameters corresponding to the first information are determined.

[0173] In some embodiments, the processing module 702 is configured to:

[0174] Based on the evaluation model, the first information and the first input information are processed to obtain the evaluation parameters corresponding to the first information; the evaluation model is used to evaluate the degree of correlation between the two types of input information.

[0175] In some embodiments, the processing module 702 is configured to:

[0176] When the evaluation parameter is less than or equal to a preset threshold, second information is generated using a second language model based on the first input information.

[0177] In some embodiments, the first input information includes initial input information, and the processing module 702 is configured to:

[0178] Determine the prompt information corresponding to the initial input information. The prompt information is used to characterize the perspective of answering the question described by the initial input information.

[0179] The prompt message is concatenated with the initial input message to obtain the updated initial input message.

[0180] Regarding the apparatus in the above embodiments, the specific manner in which each module performs its operation has been described in detail in the embodiments related to the method, and will not be elaborated upon here.

[0181] This disclosure also provides a terminal, including: a processor; and a memory for storing processor-executable instructions; wherein the processor is configured to execute the image processing method described above.

[0182] Figure 8 This is a block diagram of a terminal 800 according to an exemplary embodiment.

[0183] Reference Figure 8 Terminal 800 may include one or more of the following components: processing component 802, memory 804, power supply component 806, multimedia component 808, audio component 810, input / output (I / O) interface 812, sensor component 814, and communication component 816.

[0184] Processing component 802 typically controls the overall operation of terminal 800, such as operations associated with display, telephone calls, data communication, camera operation, and recording. Processing component 802 may include one or more processors 820 to execute instructions to complete all or part of the steps of the methods described above. Furthermore, processing component 802 may include one or more modules to facilitate interaction between processing component 802 and other components. For example, processing component 802 may include a multimedia module to facilitate interaction between multimedia component 808 and processing component 802.

[0185] Memory 804 is configured to store various types of data to support operation on terminal 800. Examples of this data include instructions for any application or method operating on terminal 800, contact data, phonebook data, messages, pictures, videos, etc. Memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0186] Power supply component 806 provides power to various components of terminal 800. Power supply component 806 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to terminal 800.

[0187] Multimedia component 808 includes a screen that provides an output interface between the terminal 800 and the user. In some embodiments, the screen may include a liquid crystal display (LCD) and a touch panel (TP). If the screen includes a touch panel, the screen may be implemented as a touchscreen to receive input signals from the user. The touch panel includes one or more touch sensors to sense touches, swipes, and gestures on the touch panel. The touch sensors may sense not only the boundaries of the touch or swipe action but also the duration and pressure associated with the touch or swipe operation. In some embodiments, multimedia component 808 includes a front-facing camera and / or a rear-facing camera. When the terminal 800 is in an operating mode, such as a shooting mode or a video mode, the front-facing camera and / or the rear-facing camera may receive external multimedia data. Each front-facing camera and rear-facing camera may be a fixed optical lens system or have focal length and optical zoom capabilities.

[0188] Audio component 810 is configured to output and / or input audio signals. For example, audio component 810 includes a microphone (MIC) configured to receive external audio signals when terminal 800 is in an operating mode, such as call mode, recording mode, and voice recognition mode. The received audio signals may be further stored in memory 804 or transmitted via communication component 816. In some embodiments, audio component 810 also includes a speaker for outputting audio signals.

[0189] I / O interface 812 provides an interface between processing component 802 and peripheral interface modules, such as keyboards, click wheels, buttons, etc. These buttons may include, but are not limited to, home buttons, volume buttons, power buttons, and lock buttons.

[0190] Sensor assembly 814 includes one or more sensors for providing state assessments of various aspects of terminal 800. For example, sensor assembly 814 can detect the on / off state of terminal 800, the relative positioning of components such as the display and keypad of terminal 800, changes in position of terminal 800 or one of its components, the presence or absence of user contact with terminal 800, orientation or acceleration / deceleration of terminal 800, and temperature changes of terminal 800. Sensor assembly 814 may include a proximity sensor configured to detect the presence of nearby objects without any physical contact. Sensor assembly 814 may also include a light sensor, such as a CMOS or CCD image sensor, for use in imaging applications. In some embodiments, sensor assembly 814 may also include an accelerometer, gyroscope, magnetometer, pressure sensor, or temperature sensor.

[0191] Communication component 816 is configured to facilitate wired or wireless communication between terminal 800 and other devices. Terminal 800 can access wireless networks based on communication standards, such as WiFi, 2G, or 3G, or combinations thereof. In one exemplary embodiment, communication component 816 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In one exemplary embodiment, communication component 816 also includes a near-field communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on radio frequency identification (RFID) technology, Infrared Data Association (IrDA) technology, ultra-wideband (UWB) technology, Bluetooth (BT) technology, and other technologies.

[0192] In an exemplary embodiment, terminal 800 may be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the methods described above.

[0193] In an exemplary embodiment, a non-transitory computer-readable storage medium including instructions is also provided, such as a memory 804 including instructions, which can be executed by a processor 820 of a terminal 800 to perform the above-described method. For example, the non-transitory computer-readable storage medium may be a ROM, random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0194] This disclosure also provides a non-transitory computer-readable storage medium, wherein when the instructions in the storage medium are executed by a terminal's processor, the terminal is able to execute the information generation method provided in the exemplary embodiments of this disclosure.

[0195] This disclosure also provides a cloud server, including: a processor; a memory for storing processor-executable instructions; wherein the processor is configured to execute the information generation method described above.

[0196] This disclosure also provides a non-transitory computer-readable storage medium, which, when the instructions in the storage medium are executed by a processor of a cloud server, enables the cloud server to perform the information generation method provided in the exemplary embodiments of this disclosure.

[0197] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and spirit of this disclosure are indicated by the following claims.

[0198] It should be understood that this disclosure is not limited to the precise structures described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this disclosure is limited only by the appended claims.< / answer> < / prompt> < / answer> < / prompt>

Claims

1. An information generation method, characterized in that, include: Based on the first input information, the first information is generated using the first language model; The first language model is the large language model configured in the terminal; Obtain the evaluation parameters corresponding to the first information; the evaluation parameters are determined by the cloud server and are used to characterize the degree of correlation between the first information and the first input information. When the evaluation parameter is greater than a preset threshold, the first information is determined as the first output information corresponding to the first input information.

2. The information generation method according to claim 1, characterized in that, The method further includes: When the evaluation parameter is less than or equal to the preset threshold, second information is obtained; the second information is information generated by the second language model based on the first input information, and the second language model is the large language model configured in the cloud server. The second information is determined as the first output information corresponding to the first input information.

3. The information generation method according to claim 2, characterized in that, The first input information includes initial input information and historical output information, wherein the historical output information includes output information generated based on the initial input information prior to the first information; the method further includes: If no target identifier is detected from the first output information, the initial input information, the historical output information, and the first output information are determined as the second input information; the target identifier is used to indicate the end of the information generation process. Based on the second input information, determine the second output information corresponding to the second input information until the target identifier is detected from the output information.

4. The information generation method according to claim 3, characterized in that, The method further includes: If the target identifier is detected from the first output information, the historical output information and the first output information are determined as the target output information corresponding to the initial input information.

5. The information generation method according to claim 3, characterized in that, The method further includes: Obtain updated initial input information; the updated initial input information includes the initial input information and the prompt information corresponding to the initial input information, the prompt information being used to characterize the angle of answering the question described by the initial input information; Based on the updated initial input information, the first input information is determined.

6. An information generation method, characterized in that, include: Obtain first information, which is information generated by a first language model based on first input information, and the first language model is a large language model configured in the terminal; The first information and the first input information are processed to obtain the evaluation parameters corresponding to the first information; the evaluation parameters are used to characterize the degree of correlation between the first information and the first input information.

7. The information generation method according to claim 6, characterized in that, The first information includes multiple semantic units. The process of processing the first information and the first input information to obtain the evaluation parameters corresponding to the first information includes: The first input information and each semantic unit are processed to obtain the evaluation parameters of each semantic unit; The evaluation parameters corresponding to the first information are determined based on the evaluation parameters of each semantic unit.

8. The information generation method according to claim 6 or 7, characterized in that, The process of processing the first information and the first input information to obtain the evaluation parameters corresponding to the first information includes: Based on the evaluation model, the first information and the first input information are processed to obtain the evaluation parameters corresponding to the first information; the evaluation model is used to evaluate the degree of correlation between the two types of input information.

9. The information generation method according to claim 7, characterized in that, The method further includes: When the evaluation parameter is less than or equal to a preset threshold, second information is generated using the second language model based on the first input information.

10. The information generation method according to claim 6, characterized in that, The first input information includes initial input information, and the method further includes: Determine prompt information corresponding to the initial input information, wherein the prompt information is used to characterize the angle of answering the question described by the initial input information; The prompt information is concatenated with the initial input information to obtain the updated initial input information.

11. An information generation device, characterized in that, include: The generation module is configured to generate first information based on the first input information and using a first language model. The first language model is the large language model configured in the terminal; The first acquisition module is configured to acquire the evaluation parameters corresponding to the first information; the evaluation parameters are determined by the cloud server and are used to characterize the degree of correlation between the first information and the first input information. The determination module is configured to determine the first information as the first output information corresponding to the first input information when the evaluation parameter is greater than a preset threshold.

12. An information generation device, characterized in that, include: The second acquisition module is configured to acquire first information, which is information generated by a first language model based on the first input information, and the first language model is a large language model configured in the terminal. The processing module is configured to process the first information and the first input information to obtain the evaluation parameters corresponding to the first information; the evaluation parameters are used to characterize the degree of correlation between the first information and the first input information.

13. A terminal, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to perform the information generation method as described in any one of claims 1-5.

14. A cloud server, characterized in that, include: processor; Memory used to store processor-executable instructions; The processor is configured to perform the information generation method as described in any one of claims 6-10.

15. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the terminal, the terminal is able to perform the information generation method as described in any one of claims 1-5.

16. A non-transitory computer-readable storage medium, characterized in that, When the instructions in the storage medium are executed by the processor of the cloud server, the cloud server is able to perform the information generation method as described in any one of claims 6-10.

Images